Analyzing COVID-19 on Online Social Media: Trends, Sentiments and Emotions Xiaoya Li♣, Mingxin Zhou♣, Jiawei Wu♣, Arianna Yuan, Fei Wu♠ and Jiwei Li♣ ♠ Department of Computer Science and Technology, Zhejiang University Computer Science Department, Stanford University ♣ Shannon.AI {xiaoya_li, mingxin_zhou, jiawei_wu, jiwei_li}@shannonai.com [email protected], [email protected] Abstract—At the time of writing, the ongoing pandemic of coron- People constantly post about the pandemic on social media avirus disease (COVID-19) has caused severe impacts on society, such as Twitter, Weibo and Facebook. They express their atti- economy and people’s daily lives. People constantly express their tudes and feelings regarding various aspects of the pandemic, opinions on various aspects of the pandemic on social media, making user-generated content an important source for understanding public such as the medical treatments, public policy, their worry, etc. emotions and concerns. Therefore, user-generated content on social media provides an important source for understanding public emotions and In this paper, we perform a comprehensive analysis on the affective trajectories of the American people and the Chinese people based on concerns. Twitter and Weibo posts between January 20th, 2020 and May 11th In this paper, we provide a comprehensive analysis on the 2020. Specifically, by identifying people’s sentiments, emotions (i.e., anger, disgust, fear, happiness, sadness, surprise) and the emotional affective trajectories of American people and Chinese people triggers (e.g., what a user is angry/sad about) we are able to depict the based on Twitter and Weibo posts between January 20th, dynamics of public affect in the time of COVID-19. By contrasting 2020 and May 11th 2020. We identify fine-grained emotions two very different countries, China and the Unites States, we reveal (including anger, disgust, fear, happiness, sadness, surprise) sharp differences in people’s views on COVID-19 in different cultures. expressed on social media based on the user-generated content. Our study provides a computational approach to unveiling public emotions and concerns on the pandemic in real-time, which would Additionally, we build NLP taggers to extract the triggers of potentially help policy-makers better understand people’s need and different emotions, e.g., why people are angry or surprised, thus make optimal policy. what they are worried, etc. We also contrast public emotions between China and the Unites States, revealing sharp differ- ences in public reactions towards COVID-19 related issues in I. INTRODUCTION different countries. The emergence of COVID-19 in early 2020 and its subsequent By tracking the change of public sentiment and emotion outbreak have affected and changed the world dramatically. over time, our work sheds light on the evolution of public According to the World Health Organization (WHO), by mid- attitude towards this global crisis. This work contributes to May 2020, the number of confirmed COVID-19 cases has the growing body of research on social media content in the reached 5 millions with death toll over 300,000 world wide. time of COVID-19. Our study provides a way to extracting arXiv:2005.14464v3 [cs.SI] 5 Jun 2020 Several mandatory rules have been introduced by the govern- public emotion towards the pandemic in real-time, and could ment to prevent the spread of the coronavirus, such as social potentially lead to better decision-making and the development distancing, bans on social gatherings, store closures and school of wiser interventions to fight this global crisis. closures. Despite their positive effects on slowing the spread The rest of this paper is organized as follows: we briefly go of the pandemaic, they neverthless caused severe impacts on through some related work in Section 2. We then present the the society, the economy and people’s everyday life. There have been anti-lockdown and anti-social-distancing protests in analyses on topic trends in Weibo and Twitter (section 3), the extracted emotion trajectories (section 4) and triggers of those many places around the world. Given these difficult situations, emotions (section 5). We finally conclude this paper in Section it is crucial for policy-makers to understand people’s opinions toward the pandemic so that they can (1) balance the concerns 6. of stoping the pandemic on the one hand and keeping people in good spirits on the other hand and (2) anticipate people’s II. RELATED WORK reactions to certain events and policy so that the policy- makers can prepare in advance. More generally, a close look A. Analyses on Social Media Content about COVID-19 at the public affect during the time of COVID-19 could help us understand people’s reaction and thoughts in the face of At the time of writing, analyses on people’s discussions and extreme crisis, which sheds light on humanity in moments of behaviors on social media in the context of COVID-19 has darkness. attracted increasing attention. [1] analyzed tweets concerning COVID-19 on Twitter by selecting important 1-grams based real-world outcomes such as economic trends [24], [28], [29], on rank-turbulence divergence and compare languages used [30], stock market [31], [32], influenza outbreak [33], and in early 2020 with the ones used a year ago. The authors political events [34], [35], [36], [37]. observed the first peak of public attention to COVID-19 around January 2020 with the first wave of infections in China, and the second peak later when the outbreak hit many western III. GENERAL TRENDS FOR COVID-19 RELATED POSTS countries. [2] released the first COVID-19 Twitter dataset. [3] provided a ground truth corpus by annotating 5,000 texts In this section, we present the general trends for COVID19- (2,500 short + 2,500 long texts) in UK and showed people’s related posts on Twitter and Weibo. We first present the semi- worries about their families and economic situations. [4] supervised models we used to detect COVID-19 related tweets. viewed emotions and sentiments on social media as indicators Next we present the analysis on the topic trends on the two of mental health issues, which result from self-quarantining social media platforms. and social isolation. [5] revealed increasing amount of hateful speech and conspiracy theories towards specific ethnic groups such as Chinese on Twitter and 4chan’s. Other researchers A. Retrieving COVID-19 Related Posts started looking at the spread of misinformation on social media For Twitter, we first obtained 1% of all tweets that are written [6], [7]. [8] provide an in-depth analysis on the diffusion of in English and published within the time period between misinformation concerning COVID-19 on five different social January 20th, 2020 and May 11th 2020. The next step is to platforms. select tweets related to COVID-19. The simplest way, as in [2], [7], is to use a handcrafted keyword list to obtain tweets containing words found in the list. However, this method leads B. Analyses of Emotions and Sentiments on Social Media to lower values in both precision and recall: for precision, user- generated content that contains the mention of a keyword is not Discrete Emotion Theory [9], [10], [11] think that all humans necessarily related to COVID-19. For example, the keyword have an innate set of distinct basic emotions. Paul Ekman and list used in [2] include the word China, and it is not suprising his colleagues [12] proposed that the six basic emotions of that a big proportion of the posts containing “China" is not humans are anger, disgust, fear, happiness, sadness, and sur- related to COVID-19; for recall, keywords for COVID-19 can prise. Ekman explains that different emotions have particular change over time and might be missing in the keyword list. characteristics expressed in varying degrees. Researchers have debated over the exact categories of discreate emotions. For To tackle this issue, we adopt a bootstrapping approach. The instance, [13] proposed eight classes for emotions including bootstrapping approach is related to previous work on semi- love, mirth, sorrow, anger, energy, terror, disgust and astonish- supervised data harvesting methods [38], [39], [40], in which ment. we build a model that recursively uses seed examples to extract patterns, which are then used to harvest new examples. Automatically detecting sentiments and emotions in text is a Those new examples are further used as new seeds to get crucial problem in NLP and there has been a large body of new patterns. To be specific, we first obtained a starting seed work on annotating texts based on sentiments and building keyword list by (1) ranking words based on tf-idf scores machine tools to automatically identify emotions and senti- from eight COVID-19 related wikipedia articles; (2) manually ments [14], [15], [16], [17]. [18] created the first annotated examining the ranked word list, removing those words that are dataset for four classes of emotions, anger, fear, joy, and apparently not COVID-19 related, and use the top 100 words sadness, in which each text is annotated with not only a label in the remaining items. Then we retrieved tweets with the of emotion category, but also the intensity of the emotion mention of those keywords. Next, we randomly sampled 1,000 expressed based on the Best–Worst Scaling (BWS) technique tweets from the collection and manually labeled them as either [19]. A follow-up work by [20] created a more comprehen- COVID-19 related or not. The labeled dataset is split into the sively annotated dataset from tweets in English, Arabic, and training, development and test sets with ratio 8:1:1. A binary Spanish. The dataset covers five different sub-tasks including classification model is trained on the labeled dataset to classify emotion classification, emotion intensity regression, emotion whether a post with the mention of COVID-related keywords intensity ordinal classification, valence regression and valence is actually COVID-related.